Homelander

@AnthropicAI Sam Altman now :

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. anthropic.com/glasswing

@AnthropicAI x.com/Kutral_eswar/s…

Full report here: anthropic.com/glasswing

Seneca said: "There is no favorable wind for the sailor who doesn't know where he's going." The destination is clear. Defenders move first — or everyone loses. The butterfly's wings are transparent for a reason. So are the bugs hiding in your software. 🦋 RT if this should be heard.

An AI just found a 27-year-old security hole in one of the world's most "unhackable" operating systems. Nobody caught it for three decades. The AI found it in hours. Alone. No human help. We are no longer in the era of human-speed cybersecurity. 🧵

@bridgemindai Yupp

@Kutral_eswar Do you have an account?

GLM 5.1 just dropped on OpenRouter. $1.40/$4.40 per million tokens. 202K context. Released today. The model that claims to work independently on a single task for over 8 hours. Autonomously planning, executing, and self-improving. That's a bold claim. Time to prove it. BridgeBench results coming soon. bridgebench.ai

@bridgemindai Thread full of replies from comet 😂

The truth is finally out. Anthropic just emailed every Claude subscriber. The rate limits weren't a bug. Third party tools like OpenClaw were putting "outsized strain" on their systems. Anthropic's fix? Cut them off. Starting April 4, third-party harnesses no longer get your subscription limits. Pay-as-you-go only. To make up for it, every subscriber gets a one time credit equal to their monthly subscription. I'm getting $200 in extra usage. This is huge. If this is what was killing Claude Code rate limits for Max plan users, tomorrow should feel like a completely different product. I'll be testing Claude Opus 4.6 all day and reporting back. Stay tuned.

@geminicli Yeah you lost most of your paying customers

Gemini CLI v0.36.0 Release Notes📝 We are aware some users are seeing long delays before the agent responds or high error rates. We apologize for this issue and are actively working on improving the situation and will update folks once we make progress. • Subagents are now enabled (give them a try)🤖 • Refreshed UI ✨ • Faster startup times ⚡️ Read full notes below 👇

@hcompany_ai @bridgemindai Checkout

Holo3 is here 🚀. Today, we're launching Holo3: our new series of frontier computer-use models. 78.9% on OSWorld-Verified. That puts us ahead of GPT-5.4 and Opus 4.6, at one-tenth of the cost. Weights on Hugging Face. API is live. Test it now! #Holo3 #OpenSource #ComputerUse #OSWorld #AI #AgenticAI

@bridgemindai Use it for Bridge Cli

The entire Claude Code source code just leaked. All of it. 57MB source map file. 2,304 files. 785K main.tsx. 103 commands. 331 utils. 146 components. 87 hooks. 45 tools. Anthropic accidentally shipped full source maps in their npm package. Anyone who installed Claude Code had the entire unminified TypeScript codebase sitting on their machine. Hidden slash commands. Unreleased features. Internal employee god mode. Future model references. Anti-distillation defenses. All exposed. This is the biggest accidental source code leak in AI history.

@0xSero Give to someone who needs it. Like me👀

Giving away 5 Opencode Go subs Winners selected randomly from comments in 24 hours.

↓ Grab v0.0.7 here: github.com/Kutral/VoxDrop… .exe and .msi both available. ~3–4MB. No bloat. If you've been sleeping on VoxDrop — this is the build to try.

VoxDrop v0.0.7 is live 🚀 Two big upgrades this drop: 🔊 Audio Earcons — you now *hear* VoxDrop work. Crisp sweep when you press the hotkey. Soft chime when text pastes. Operate it entirely by feel. No screen watching needed. ⚡ Pasting is now instant. Replaced keystroke simulation with direct Win32 clipboard injection. Lightning-fast. And your dictated text never touches Clipboard History. Free. Open source. Built for speed & privacy. ↓ Download link in reply

@bridgemindai x.com/Kutral_eswar/s…

Gemini CLI with Gemini 3.1 Pro is one of the worst vibe coding tools on the market. The model itself is a solid baseline. #1 on BridgeBench Hallucination. Decent coding intelligence. But the tooling around it is terrible. Crashes mid-session. Loses context. Can't complete multi-file tasks reliably. I cancelled my $250/month Google AI Ultra plan for a reason. Google has a competitive model wrapped in unusable infrastructure. Claude Code with Claude Opus 4.6 and Codex with GPT 5.4 are years ahead on developer experience. Who else is having problems with Gemini CLI?

@bridgebench I meant overall coding performance bench.

@Kutral_eswar Already live on bridgebench.ai. Check out BridgeBench Code.

GLM 5.1 improved on hallucination over GLM 5. Still not enough. GLM 5 fabricated 38.6% of the time. GLM 5.1 brought it down to 34.1%. Better. But still the worst frontier-class model on BridgeBench Hallucination. Z.ai closed the intelligence gap with Claude Opus 4.6 and GPT 5.4 on coding benchmarks. They haven't closed the trust gap. A model that makes things up a third of the time isn't ready for production codebases. bridgebench.ai

@geminicli WHY should we use this ? @geminicli

Gemini CLI v0.35.2 Release Notes📝 • Fully customize your keybindings ⌨️ • Vim mode got some love (/vim) 🎮 • Git worktree support is now in preview 🌲 Read full notes below 👇